R/bovar.R

In quint: Qualitative Interaction Trees

bovar<-function(y,Xmat,tr,gm,dmatsg,dmatsel,parents,parvec,w,nsplit,crit=crit)  {
  #selects best observed splitting candidate
  #create matrix to keep the highest value of the criterion
  critmax<-matrix(0,nrow=dim(Xmat)[2],ncol=6)
  critmax[,1]<-1:ncol(Xmat)
  #only admissible rows of the designmatrix for this node
  dmats<-dmatsg[dmatsel==1,]
  index<-1:nrow(dmatsg)
  #split only if there are more than 1 persons in the node
  if(sum(gm)>1){
    critmax[,c(2:6)]<-t(sapply(1:dim(Xmat)[2],function(kk,y,Xmat,tr,gm,dmats,parents,parvec,w,nsplit,crit){
      bos(y,x=Xmat[,kk],tr,gm,dmats,parents,parvec,w,nsplit,crit=crit)},
      Xmat=Xmat,gm=gm,y=y,tr=tr,parents=parents,nsplit=nsplit,w=w,parvec=parvec,dmats=dmats,crit=crit) )
    #which predictor is the best spllting candidate for this split?
    bestrow<-which(critmax[,4]==max(critmax[,4]))[1] }
  else{bestrow<-1}
  #rowstarnew is rownumber in the general designmatrix dmatsg
  if( critmax[bestrow,3]!=0){
    critmax[bestrow,3]<-index[dmatsel==1][critmax[bestrow,3]] }
  return(critmax[bestrow,])}